Upload model.safetensors with huggingface_hub

Clean up gitignore for HF push

Longer DNA prompt

Show subtractions of HF and checkpoint key sets in inspection script

Fix PT to safetensors conversion

Add Evo 2 checkpoint download script to gitignore

Downgrade max sample length in custom config to avoid CUDA OOM error

Remove more custom device management

Update inspect keys script

Remove more manual device management logic from model.py

Replace TELinear with nn.Linear

Fix a relative import

Add script to compare keys of HF model and checkpoint

Uncomment layer assignment

Disable Flash Attention (for now)

Fix tokenization decoding error

Remove device management from StripedHyena forward pass for x

Change past_key_values indexing from hyena (inherited from Together's Evo 1 HF code) to hcl, hcm and hcs

Add use_cache argument in config definition

Change generate() arguments to the format expected by Evo 2

Fix device management bug

PT to safetensors

Fix JSON dump bug

Add logs to gitignore

Fix relative import of attention

Local test scripts

Custom-written HF glue; config JSON is translated from Evo 2 YAML and Python is based on Together's HF port code

Rich logging from vortex

Gitignore

Copy rotary from vortex; drop-in replace vortex.ops apply_rotary with flash_attn's apply_rotary

Copy Python verbatim from vortex

License and readme

Add tokenizer files verbatim from Together's Evo 1 HF

initial commit